Full covariance state duration modeling for HMM-based speech synthesis | IEEE Conference Publication | IEEE Xplore